Well, I think it depends. Both could be valid strategies. When generating 
surfaces from multi-label segmentations, there's always the question how 
accurate it have to be, like, do you actually want to model the 
voxel-boundaries as polygons to get blocky surfaces, or would you expect smooth 
surfaces. While smooth surfaces are okay for many classic segmentations, it is 
probably impossible to find a single best approach for multi-label 
segmentations (i. e. should surfaces of adjacent labels also "touch" each other 
and how). Instead of multiple surfaces, it should also be possible to create a 
single surface and using vtk vertex colors for the different labels.

If you do not really need to store surfaces but instead are happy with just a 
3-d visualization, volume rendering seems to be a nice alternative. When we 
upgraded to the latest VTK with the OpenGL2 backend a few months ago in our 
master branch, the volume mapper was also upgraded (basically to the latest VTK 
volume mapper). It may already have all the functionality needed to retrieve a 
nicely visualized volume out of an image representing multiple labels. Didn't 
look into it so far, though.

I would implement "visualization nodes" as invisible helper nodes as children 
to their corresponding image node, just like the segmentation uses helper nodes 
for temporary feedback contour nodes. 


-----Original Message-----
From: Clarkson, Matt [mailto:m.clark...@ucl.ac.uk] 
Sent: Donnerstag, 9. November 2017 16:42
To: Dinkelacker, Stefan
Cc: MITK
Subject: Re: Multi-Label Visualisations

Thanks Stefan,

it might take a bit of work, and planning.

For example, if you have an image with automatically generated labels, say up 
to 256 from some automatic segmentation, and then you want to visualise each 
region as a different colour.

If you do surface rendering, you’d need to generate 256 new surfaces. But would 
these be hidden, and just linked to the original image, as parent-child 
relationship?
Generating them might take some time, and what sort of colour maps would be 
people want?
Then if you delete the image, do you delete all the surfaces?

Or would volume rendering suffice? i.e. generate a LUT with 256 colours, and 
have some ray-tracing function that doesnt interpolate and makes things look 
fully opaque.

Im not sure at this stage.

M


> On 9 Nov 2017, at 14:42, Dinkelacker, Stefan 
> <s.dinkelac...@dkfz-heidelberg.de> wrote:
> 
> Hi Matt,
> 
> as far as I know there is currently no quick way to visualize 
> multi-label segmentations in 3-d other than doing it by a custom 
> ITK/VTK pipeline. As the multi-label segmentation still has prototype 
> character in some aspects, we will discuss the segmentation in detail 
> next week, to address these kinds of issues in near future. :-)
> 
> Best,
> Stefan
> 
> -----Original Message-----
> From: Clarkson, Matt [mailto:m.clark...@ucl.ac.uk]
> Sent: Donnerstag, 9. November 2017 13:57
> To: MITK
> Subject: [mitk-users] Multi-Label Visualisations
> 
> Hi there,
> 
> Is there a way in MITK of quickly/automatically doing 3D surface or volume 
> rendering of multi-label segmentations?
> If so, which version of MITK is it available in, as we are a bit behind the 
> curve at the moment.
> 
> Thanks
> 
> M
> 
> 
> ----------------------------------------------------------------------
> -------- Check out the vibrant tech community on one of the world's 
> most engaging tech sites, Slashdot.org! http://sdm.link/slashdot 
> _______________________________________________
> mitk-users mailing list
> mitk-users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/mitk-users

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
mitk-users mailing list
mitk-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mitk-users

Reply via email to